- 83 views
- 84 downloads
Learning Individual Readmission-Free Survival Distributions using Longitudinal Medical Events
-
- Author / Creator
- Davis, Sarah (Sacha) Maren
-
The rate of 30-day hospital readmission is a common measurement of hospital quality, which can affect the funding a hospital receives. Over a quarter of readmissions are estimated to be preventable with adequate interventions, but these interventions are themselves costly. For this reason, many projects have attempted to determine which individuals are at a high risk of readmission, and thus whose prognosis may improve with further testing and treatment. There are two common approaches to this prediction problem. (1) Formulate risk indices, such as the LACE score. These are common in a hospital setting; however, the simplicity often leads to poor predictive performance. (2) Use machine learning to transform a set of hand-selected features into the probability of readmission at a single future time-point. Unfortunately, feature engineering is time-consuming, and a physician may care about predictions at time points other than 30 days. Both approaches rely heavily on domain knowledge. In this thesis, I use Neural Multi-Task Logistic Regression (N-MTLR) to model all-cause readmission-free survival as a function of time. N-MTLR, despite producing probability predictions for all future time-points, out-performs XGBoost and Deep Learning approaches trained specifically to predict readmissions at 30 days (AUROC 0.821±0.004 (Std.Dev) versus 0.814±0.003 and 0.810±0.005). Further, I show that N-MTLR, augmented with a sequence model, can learn a patient’s representation directly from their history of medical codes, predicting 30-day all-cause readmission with an AUROC of 0.846±0.003 using only sequences of historical medical codes as input. This approach significantly outperforms the LACE baseline of 0.659±0.001. These results demonstrate the merit of medical code sequences to represent a patient’s entire past, and N-MLTR to model a patient’s entire future.
-
- Graduation date
- Fall 2023
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.